Study On Fault Diagnosis Of Rotating Machinery Based On Kurtosis Spectral And High-Order Cumulants

In this dissertation bearing and gear fault signal was analysed from two aspects:feature extraction and pattern recognition. In order to determine filter parameters, a feature extraction method based on kurtosis spectral index was studied, and matched filter based on higher-order cumulants was constructed, This paper also proposed a feature extraction method based on 1.5 dimension spectral and combined it with the dual channel data fusion method, the research of the composite fault diagnosis method was carried out; Shuffled Frog Leaping Algorithm was applied to fault diagnosis. Research contents are as follows:(1)On the premise of filter parameters selection lacks theory basis, the spectral kurtosis was calculated hierarchically though Fast Kurtogram and the signal was filtered at the frequency band whose spectral kurtosis was maximum and realized the extraction of resonance frequency band. In addition, center frequency and bandwidth were selected though iterative refinement called Protrugram, which extracted the signal feature though the minimum filter bandwidth.(2)Studied the matched filter based on spectral kurtosis index due to the optimal frequency band could not be determined in each case by Fast Kurtogram. Matched filter was obtained by calculating the eigenvectors corresponding to the largest eigenvalue, and its performance was improved through three-order cumulants. Then, the superiority of matched filter was verified with the same data of fault bearing.(3)The fault feature extraction method was studied based on high-order cumulants.The Fourier transform of three-order cumulants diagonal slice was used to obtain the 1.5 dimension spectrum,which was insensitive to Gaussian noise characteristics.Then, the dual channel data fusion method is combined with the 1.5 dimensional spectrum, and the identification of single fault and mixed fault of bearing is realized.(4)The Shuffled Frog Leaping Algorithm (SFLA) was used to diagnose rotating machine fault. Various kinds of conventional feature parameters was selected and constructed the kurtosis spectral entropy. The fault data of bearing and gear were analyzed though Shuffled Frog Leaping Algorithm, and the different data pattern identification is realized.